In May 2005, the Organ Procurement and Transplantation Network (OPTN) modified their lung transplantation policy from one in which potential recipients were prioritized based on the amount of time spent on the waiting list to one in which donor lungs were allocated to recipients based on a lung allocation score (LAS). Since the adoption of the LAS, studies have shown that the mortality rate for patients on the lung transplant waiting list has decreased, but resource use and geographic and gender disparities in lung allocation have increased. Additionally, while the statistical models used to construct the LAS control for patients? demographics, diagnoses, and laboratory values, they do not account for the fact that in order to receive a lung transplant, an individual must survive on the waiting list long enough for a suitable lung to become available. Since individuals who survive one year or more on the waiting list might be inherently different from individuals who die, receive a transplant, or are censored prior to one year, failure to incorporate this information can lead to inaccurate LAS predictions. Epidemiologists refer to such bias as ?survivor bias.? The goal of this research and training plan is to improve the predictive accuracy of the LAS by accounting for survivor bias so that lungs are allocated to the appropriate patients in the appropriate order. This goal will be accomplished via three aims.
In Aim 1, we will use advanced causal inference methods, such as inverse probability weighting, to quantify the bias imposed by the fact that individuals who survive one year or more on the waiting list might be inherently different from individuals who die, receive transplant, or are censored prior to one year. Such methods rely on the ?potential outcomes? framework to ?map? the survival probabilities obtained among the post-transplant group back to the full waiting list population.
In Aim 2, we will compare the predictive accuracy of the model developed in Aim 1 to the original LAS. Specifically, we will look at model calibration, discrimination, and computational efficiency. Finally, in Aim 3, we will use qualitative research methods to begin to 1) understand how physicians use the current LAS to make treatment decisions in lung disease care, 2) determine whether the modifications to the LAS from Aim 1 alters their decision-making process, and 3) provide the applicant with the training necessary to conduct future qualitative research. This innovative approach will allow the transplant community to better understand how the current LAS is used in clinical practice and how survivor bias influences physicians? estimates of the benefit of transplant. Findings from this study can also be applied in other areas of medicine which rely upon prediction models, such as cardiology and intensive care. The accompanying training plan consists of both didactic and experiential learning opportunities, and will enable the applicant to develop the skills and experience necessary to become an independent investigator at the intersection of epidemiology and biostatistics.
The current approach for allocating donor lungs to potential recipients ? namely, the Lung Allocation Score (LAS) ? does not account for the fact that, in order to receive a lung transplant, an individual must survive on the waiting list long enough for a suitable lung to become available. This can lead to biased prediction estimates that could undermine the equity and utility of the lung allocation process and affect clinicians? decision making. Using publicly available quantitative health outcomes data collected by the United Network for Organ Sharing (UNOS), along with qualitative interview data of lung transplant surgeons and pulmonologists, this proposal will build upon methods available to account for such bias, and enable the lung transplant community to move toward more comprehensive assessments of transplant benefit.
